Landing: f0241f047bb0
Project / Subsystem
gcc / gcc/fortran
Date
2026-08-01
Author
Thomas Koenig
Commit
f0241f047bb0858964119e3aaad2bbd61e7d757f
Source
github
Perf win
Yes
Breaking
No
All attributes
- project
- gcc
- subsystem
- gcc/fortran
- patch_id
- —
- commit_hash
- f0241f047bb0858964119e3aaad2bbd61e7d757f
- source_type
- github
- headline
- Inline MATMUL(TRANSPOSE(A),B) for rank 1 B.
- tldr
- Optimized inlining of matrix multiplication involving a transposed matrix and a rank-1 matrix.
- author
- Thomas Koenig
- outcome
- committed
- performance_win
- true
- breaking_change
- false
- series_id
- —
- series_parts
- []
- tags
-
- • fortran
- • optimization
- • matmul
- discussion_id_link
- —
- bugzilla_pr
- —
- date
- 2026-08-01T00:00:00.000Z
This commit enables the inlining of MATMUL(TRANSPOSE(A),B) when B is a rank-1 matrix. While loop ordering presents challenges due to dependencies, GCC’s unrolling capabilities allow for better code generation, especially for smaller matrices. The change modifies frontend passes to handle this specific matrix case.